vtune returns error when analysis uarch-exploration

[oleotiger]

I'm using Intel(R) oneAPI VTune(TM) Profiler 2021.1.1 Gold (build 613804).

When I wanna collect uarch-exploration with command "vtune -collect uarch-exploration -- sleep 3", I got errors like:

"vtune: Error: Cannot configure sampling event groups. The collection is terminated.
Use -knob event-config=? to see the list of available events for the target.
vtune: Collection failed."

I found that the error occurs when "pmu-collection-mode=detailed" where detailed is the default value.

"vtune -collect uarch-exploration -knob pmu-collection-mode=detailed -- sleep 3" returns the same error message.

But "vtune -collect uarch-exploration -knob pmu-collection-mode=summary -- sleep 3" works well.

How could I solve this problem?

[Kirill_U_Intel]

Hi,

Could you share bad result directory?

Kirill

[oleotiger]

OK. The detailed directory is compressed and attached.

In r017ue-bad.vtune

<?xml version="1.0" encoding="UTF-8"?>
<root>
<rdmgr>
<timestamp type="u64_t">1616029496</timestamp>
<hostname type="string">localhost.localdomain</hostname>
<os type="string">linux</os>
<product type="string">Intel® oneAPI VTune™ Profiler 2021.1.1 Gold</product>
<buildNumber type="s32_t">613804</buildNumber>
<logicalCPUCount type="s32_t">104</logicalCPUCount>
<physicalCoreCount type="s32_t">104</physicalCoreCount>
<processorPackageCount type="s32_t">4</processorPackageCount>
<CPUFrequency type="s64_t">3000000000</CPUFrequency>
<CPUFamily type="s32_t">6</CPUFamily>
<CPUModel type="s32_t">85</CPUModel>
<CPUStepping type="s32_t">11</CPUStepping>
<CPUBrandName type="string">Intel(R) Xeon(R) Platinum 8365HC CPU @ 3.00GHz</CPUBrandName>
<isa type="string">avx512</isa>
<collectionTimeBegin type="u64_t">1616029496</collectionTimeBegin>
<collectionTimeEnd type="u64_t">1616029497</collectionTimeEnd>
<variables>
<at type="string">ue</at>
</variables>
</rdmgr>
</root>

[oleotiger]

What's more, when I collect uarch metrics with command:  vtune -collect uarch-exploration -knob pmu-collection-mode=summary -target-pid=9215

Here the pid is a process of wrf.exe in openmpi application locally.

I got a weird result:

I'm sure there are memory access and L3 miss rate is almost 80% monitoring with pcm.

Why I got 100% core bound?

[Xiaojun_H_Intel]

I encountered the same issue with 2021.1 version when I was analyzing an application on a remote server.

[oleotiger]

So how to solve the problem?

Is there any way to get another version of vtune that can work correctly?

[JananiC_Intel]

Hi,

Thanks for the update.

We will check on this internally and let you know.

Regards,

Janani Chandran

[DMITRY_T_Intel]

Hi,

I escalated this issue to development. I will let you know as soon as I get any information.

Thank you!

[Dmitry_R_Intel1]

This issue should go away in the next VTune 2021.2.0 release which is currently planned to happen next week.  

[oleotiger]

Thank you for reply. Looking forward for the next version.

[DMITRY_T_Intel]

Hi,

Developers told me that the issue should be fixed in the 2021.2 release. Please try the 2021.2 release. Can I close this case?

Thanks.

[oleotiger]

community.intel.com/t5/Analyzers/vtune-returns-error-when-analysis-uarch-exploration/m-p/1265356#M20158 

Will this issue be fixed in 2021.2?

Or it's caused by my misuse of vtune?

[Dmitry_R_Intel1]

That was the issue on VTune side. You did everything correctly.

[oleotiger]

OK. I think you can close this issue.  I will keep concern on release of vtune 2021.02. Thank you for prompt reply.

[oleotiger]

Is there an exact date that vtune  2021.2  will be released?

I'm waiting for the new version for bug fixing.

[DMITRY_T_Intel]

Hi,

The 2021.2 release is planned for this week.

Thanks.

[oleotiger]

Is it delayed?

[DMITRY_T_Intel]

Hi,

Sorry, it is a little bit delayed and should be available on this week.

Thanks.